Which test is a non-parametric method for reliability?

• A. ICC
• B. Kappa
• C. Pearson
• D. Spearman

Answer: (B)

Reliability here is about how well two raters agree on categorical judgments, and doing so in a way that doesn’t assume a particular data distribution. A non-parametric method for this purpose is designed for categorical data and doesn’t rely on normality or variance assumptions.

Kappa fits this need because it measures agreement between raters beyond what would be expected by chance. It uses a contingency table of how often raters assign each category, compares observed agreement to the agreement you’d expect if raters were guessing, and yields a value between -1 and 1. A higher value indicates stronger agreement beyond chance, and it can be adapted for ordinal data with a weighted version if you care about partial agreement.

The other options aren’t non-parametric measures of reliability in this sense. Intraclass correlation (ICC) is based on variance components from ANOVA-like models and assumes continuous data with certain distributional properties, so it’s considered parametric. Pearson correlation measures linear association between two continuous scores and also relies on distributional assumptions. Spearman is non-parametric and uses ranks, so it captures monotonic relationships, but it’s still a measure of association, not a direct reliability/agreements index between raters.